CPRI over switched Ethernet Nigel Bragg – CTO 1 Introduction Background “CPRI requirements for Ethernet Fronthaul” (802.1CM call, 24-11-15) (http://www.ieee802.org/1/files/public/docs2015/cm-CPRI-requirements-1115-v01.pdf) suggests :- maximum FDV = 5 μs or 10% of the E2E latency FDV (Frame Delay Variation) == PDV, but in IEEE 802 language FDV tolerance as an attribute of transport for a CPRI stream requires mechanisms for clock recovery and buffer control at the receiver : irrespective of the value of the maximum delay tolerance; buffer sizes are modest in a CPRI context – 50 μs @ 10G < 64 KBytes Is maximum FDV = 5 μs or 10% of the E2E latency the right choice ? where maximum E2E latency is implementation-dependent but ~ 100 μs This contribution suggests an alternative way of looking at this specification issue. 2 Functional Requirement It is believed that the root functional requirement is to hold the total one-way latency between a RRH and a BBU to less than ~ 100 μs, to allow correct operation of the hybrid ARQ mechanism : discussion with CPRI required to determine the exact latency details Transport Delay emit Frame arrival jitter maximum FDV margin time readout This suggests that the parameter which ought to be specified is Total Fronthaul Delay = Transport Delay + maximum FDV + margin < 100 μs This allows a trade-off between Transport Delay and maximum FDV : in “very short reach” deployments, FDV requirements are much relaxed maximum possible reach can still be obtained when required, by the use of precision network timing techniques. 3 Multiple Fronthaul profiles seem to have merit We show a simple use case : 2.4 Gbps CPRI bandwidth per RRH (e.g. 3-sector, single 20 MHz carrier) 10 Gbps trunking, so 4 x RRH streams supported per trunk, no pre-emption, and assume 1500 Byte packets, so unit of delay = 1.2 μs may not be realistic – a deliberate “stress test” CPRI streams RRH X 10Gbps FDV contribution (delay units) HoL blocking 1 ** “race condition” 3 CPRI streams 10Gbps X CPRI streams X 1 3 BBU 1 = 3 3 = 9 12 = 14.4 μs In this scenario, an allowance of 5 μs for “margin” still leaves 80 μs (~ 16 km) available as a Transport delay budget ** the term “race” is used as in “Delays and PDV in an Ethernet Fronthaul Network” (http://www.ieee802.org/1/files/public/docs2015/cm-farkas-delay-pdv-1215-v01.pdf ) 4 Next steps ? Suggest to CPRI Cooperation that in a switched Ethernet Fronthaul environment specification of the maximum RRH BBU delay is the preferred parameter : put Transport delay + FDV into the Ethernet Transport domain, and let that domain sort it out Extend the discussion on joint calls. 5